Building performance debugging tools to detect frequency of core migrations and dTLB/iTLB misses per thread of a given process in real time as a mini-project under Centre for Cloud Computing and Big Data at PES University.

tool1 is used to detect dTLB/iTLB misses in real time for any thread ID of a running process.
tool2 is used to detect the number of core migrations in real time for any given thread ID of a running process.

Dependencies:

• top
• perf

• PID
• USER
• S
• TIME+
• P

• Run the top command
• While it's running, press f
• Select/unselect columns by pressing d
• Once done selecting/unselecting press q
• To save the changes, press shift + w

sudo apt install linux-tools-$(uname -r) linux-tools-generic

The method specified below works for tool1 and tool2:

• Open the tool1 / tool2 folder.
• Run the following command : sudo python3 server.py
• Vist 127.0.0.1:3000 in the browser
• Now enter the PID in the corresponding field
• Now click the Update threads and select the thread to be analyzed.
• Click start thread button and wait for around 5 seconds for the dynamic graph to be displayed.
• To end the process, click on End Thread.

• Give visuals for all threads of a process
• Write a shell script to automate editing the configuration of top
• Figure out how to get more granularity in data collected
• Try and use SystemTap for collecting data